When I re-synthesize a singing voice in Pyworld, the voice is grazed as shown in the picture. Is there a workaround for this? 